Machine learning query handling system

Abstract

Systems and methods for a machine learning query handling platform are described, whereby each computing node in a computer network is configured to implement a respective local prediction model that calculates an output based on input attributes passed through trained parameters of the local prediction model, whereby at least two of the computing nodes calculate different predicted outputs to the same input attributes. In an embodiment, the trained parameters of each local prediction model include a first set of parameters received from a remote server, a second set of parameters received from another interconnected computing node, and a third set of parameters based on data in a local memory. Other embodiments are also described and claimed.

The invention claimed is: 
     
       1. A method comprising:
 storing, at a first computing device in a computer network, data defining a local prediction model including: 
 a first set of trained parameter values received from a remote server in the computer network that maintains at least one temporally-dependent attribute of a plurality of perishable units, 
 a second set of trained parameter values received from a second computing device in the computer network, and 
 a third set of local parameter values generated by the first computing device; 
 applying, by the first computing device, a machine learning algorithm to compute updated parameter values of the local prediction model based at least on data associated with the first computing device; 
 receiving, by the first computing device, an updated first set of trained parameter values from the remote server; 
 applying, by the first computing device, the machine learning algorithm to compute re-trained parameter values of the local prediction model based at least on the received updated first set of trained parameter values and the data associated with the first computing device; and 
 processing, by the first computing device, a data query using the trained local prediction model, to identify at least one perishable unit and to predict a value of at least one temporal query attribute of the or each identified perishable unit. 
 
     
     
       2. The method of  claim 1 , wherein the data query is processed using the updated parameter values of the trained local prediction model and data retrieved from a memory of the first computing device. 
     
     
       3. The method of  claim 2 , wherein the data query further comprises query attributes including the data retrieved from the memory of the first computing device. 
     
     
       4. The method of  claim 3 , wherein the query attributes further include data retrieved from a plurality of disparate data sources. 
     
     
       5. The method of  claim 3 , wherein processing the data query using the updated parameters of the local prediction model further identifies a predicted query type. 
     
     
       6. The method of  claim 1 , further comprising initialising a query vector representing query attributes of the data query, and passing the query vector as input to the trained prediction model. 
     
     
       7. The method of  claim 1 , wherein the data query is received by the first computing device from a different computing device on the network. 
     
     
       8. The method of  claim 1 , further comprising providing the predicted output as feedback to a query handling controller of the first computing device. 
     
     
       9. The method of  claim 1 , further comprising transmitting, by the first computing device, the updated first set of parameters of the local prediction model to the remote server. 
     
     
       10. The method of  claim 1 , wherein the predicted value is one or more of a probabilistic, heuristic and deterministic property of the identified perishable unit. 
     
     
       11. The method of  claim 1 , wherein the third set of local parameter values are initialised, by the first computing device, with random values. 
     
     
       12. The method of  claim 1 , wherein the updated parameter values of the trained local prediction model are computed based on a training data set that includes data associated with the first computing device. 
     
     
       13. The method of  claim 12 , wherein the training data set includes metadata associated with a plurality of perishable units. 
     
     
       14. The method of  claim 13 , wherein the training data set further includes output attributes associated with metadata of the plurality of perishable units in the data set. 
     
     
       15. The method of  claim 12 , wherein the training data set includes:
 a first data subset received from the remote server; 
 a second data subset received from the second computing device; and 
 a third data subset retrieved from a local memory of the first computing device. 
 
     
     
       16. The method of  claim 1 , wherein the data associated with the first computing device includes one or more of location data, calendar data, accelerometer data, gyroscope data, user settings, third party application data, and device type. 
     
     
       17. The method of  claim 1 , wherein the third set of local parameter values define weighting values for corresponding ones of the first set of trained parameter values. 
     
     
       18. The method of  claim 1 , further comprising iteratively updating the parameters of the local prediction model based on new input data. 
     
     
       19. An apparatus comprising:
 one or more network interfaces to communicate with a remote server and other computing devices in a computer network; 
 a processor coupled to the network interfaces and operable to execute program instructions; and 
 a memory configured to store the program instructions that, when executed by the processor, configure the processor to perform a method comprising:
 storing, at a first computing device in a computer network, data defining a local prediction model including: 
 a first set of trained parameter values received from a remote server in the computer network that maintains at least one temporally-dependent attribute of a plurality of perishable units, 
 a second set of trained parameter values received from a second computing device in the computer network, and 
 a third set of local parameter values generated by the first computing device; 
 applying, by the first computing device, a machine learning algorithm to compute updated parameter values of the local prediction model based at least on data associated with the first computing device; 
 receiving, by the first computing device, an updated first set of trained parameter values from the remote server; 
 applying, by the first computing device, the machine learning algorithm to compute re-trained parameter values of the local prediction model based at least on the received updated first set of trained parameter values and the data associated with the first computing device; and 
 processing, by the first computing device, a data query using the trained local prediction model, to identify at least one perishable unit and to predict a value of at least one temporal query attribute of the or each identified perishable unit. 
 
 
     
     
       20. A non-transitory storage medium comprising machine readable instructions stored thereon that when executed cause a computer system to perform a method comprising:
 storing, at a first computing device in a computer network, data defining a local prediction model including: 
 a first set of trained parameter values received from a remote server in the computer network that maintains at least one temporally-dependent attribute of a plurality of perishable units, 
 a second set of trained parameter values received from a second computing device in the computer network, and 
 a third set of local parameter values generated by the first computing device; 
 applying, by the first computing device, a machine learning algorithm to compute updated parameter values of the local prediction model based at least on data associated with the first computing device; 
 receiving, by the first computing device, an updated first set of trained parameter values from the remote server; 
 applying, by the first computing device, the machine learning algorithm to compute re-trained parameter values of the local prediction model based at least on the received updated first set of trained parameter values and the data associated with the first computing device; and 
 processing, by the first computing device, a data query using the trained local prediction model, to identify at least one perishable unit and to predict a value of at least one temporal query attribute of the or each identified perishable unit.